Trustworthy AI for disease pathogenesis and precision medicine: A disease-agnostic framework from data to deployment

Artificial intelligence (AI) is rapidly reshaping biomedical research and clinical practice, yet many models that perform well on benchmarks fail to deliver mechanistic insight, robust generalization, or measurable clinical value. This review presents a disease-agnostic roadmap for trustworthy AI in disease pathogenesis and precision medicine across multi-omics, medical imaging, digital pathology, electronic health records, and longitudinal real-world data. We organize the evolving model landscape from classical machine learning to deep learning, self-supervised and foundation models, graph and knowledge-guided learning, multimodal fusion, and generative approaches, while emphasizing how methodological choices interact with data provenance, endpoint validity, label design, and clinically meaningful evaluation. To bridge prediction and biology, we synthesize interpretation strategies and propose an evidence ladder that distinguishes association from robustness, biological coherence, causal support, perturbation evidence, and clinical utility. We further distill practical principles for reliable evaluation and deployment, including leakage control, external and temporal validation, calibration, decision-curve-informed thresholding, subgroup reporting, uncertainty communication, and post-deployment monitoring. By linking technical advances to evidence standards and lifecycle governance, this review provides a reusable framework for converting AI-derived signals into testable mechanisms and clinically actionable precision medicine.